Roads on edge of landuse AOIs broken up

[nvkelso]

NYC example:

Merrick Ave North and Salisbury Park Dr are getting chunked up.

https://www.openstreetmap.org/edit?#map=16/40.7429/-73.5787

SF Bay Area example:

The boundary roads along Redwood Regional Park and Chabot Regional Park (Redwood Dr, etc) are getting chunked up. Looking at the data, looks like the road and the park boundary aren't topologically the same, but I think that might be true in ground truth (because of parcel boundaries & etc) – in any case we should allow for this.

Another SF example that's less obvious north of SF near Fairfield:
#12/38.2628/-122.1076

[nvkelso]

I expect the parts of the road inside the landuse AOI to get landuse_kind, but the part outside of the AOI should still be returned (without landuse_kind added). That the part outside is missing should be investigated.

[zerebubuth]

Looks like this has been fixed as part of tilezen/TileStache#72 (in commit tilezen/TileStache@d61f347). It needs wider verification in dev, but I was able to produce the screenshots below from my own system running master which do now show this problem any more.

It seems that the issue was that the intersection was producing a GeometryCollection containing both lines and points and we were previously checking if the actual geometry type was equal to the original and, since points|lines != lines, it was dropping them. This was changed in that commit to filter the geometry type of interest from any set returned in a GeometryCollection, which seems to have addressed the issue.

Area near Salisbury Park Dr.

Zoom 15

Zoom 16

Chabot Regional Park, zoom 14

The "dashed effect" appearance of this road doesn't look nice, in my opinion, and bloats the size of the tile as each "dash" is a separate feature. However, fixing that is a separate issue.

Suisun City, zoom 12

[nvkelso]

So this is configuration changed (in #201), no further engineering work needed? Looks like #201 still needs to be taken to dev. I'll keep this on me until I can verify that #201 fixes it.

[zerebubuth]

I thought we took #201 to dev already. I just took this screenshot from dev, which seems to show it okay. Could you check that you've cleared the cache and are looking at url: //vector.dev.mapzen.com/osm/all/{z}/{x}/{y}.topojson, please? If you're still seeing old tiles then we have a problem...

[nvkelso]

Testing more with a cleared cache, I see Chabot Regional Park in dev showing the fix (matches the screenshot above), good.

But zoom 16 around Salisbury Park Dr is still broken. But zoom 15 and zoom 17 are fine. Tested in Chrome and Safari on Mac.

z15:

z16:

z17:

Could those zoom 16 tiles be stuck on dev?

[zerebubuth]

Yup, I see the same on dev, but it's fixed when looking at my local setup. Seems like that tile hasn't generated yet. We can either wait for it to regenerate, or try to fiddle with the stack to make it generate?

[rmarianski]

I was fiddling with dev earlier so it might not have gotten to this tile. I just re-enqueued zoom 16 tiles in dev. Does it look ok now?

[nvkelso]

LGTM